Modified Principal Component Analysis in sliding windowed fMRI data

The current available methods for fMRI data decomposition are mostly done as a whole ignoring the possibility of important information variability over time. While some studies applied sliding windows to the data, the independence and continuity of the data decomposition is not being taken into account. Our proposed method, the modified Principal Component Analysis (PCA) tries to force the basis vectors to be as orthogonal and independent as possible. We showed that the proposed method maintains low dot products regardless how big is the window size and overlapping window. In comparison to the classical PCA and Independent Component Analysis (ICA) methods, the method yielded the lowest dot products among them. Using this method also provided some insights on the changes in the fMRI data where there could be possibility of brain synchronizing when most of the brain regions are highly correlated.